Dissertation > Excellent graduate degree dissertation topics show

Image registration techniques based on the Scale Invariant Feature Transform (SIFT)

Author: ZhouFeng
Tutor: LiuHui
School: Kunming University of Science and Technology
Course: Computer Software and Theory
Keywords: SIFT algorithm Harris corner detection operator Image registration Scale space
CLC: TP391.41
Type: Master's thesis
Year: 2010
Downloads: 471
Quote: 2
Read: Download Dissertation

Abstract


Image registration is to match two or more images acquired at different angles at different times from different sensors or the same sensor, the process is a necessary precondition to solve the problem of image fusion, image mosaic, and super-resolution image reconstruction. In practical applications, due to the complex imaging distortion between images exist, the existing registration techniques is difficult to adapt the image has a larger scale of the alignment scaling, rotation, perspective, changes in the image, and the complexity of the algorithm is large, the accuracy of the alignment is not very high. Image registration techniques has been a very difficult subject, the study image registration techniques has important theoretical significance and practical value. Do a brief introduction to the theory and background of the image registration, image registration methods are reviewed, the hotspot of image registration techniques and trends. Existing registration feature point-based registration has high registration accuracy of disturbed small noise outside influence, and widely applicable advantages, registration is now mainstream. This article from the feature-based alignment is the main starting point for the registration of the currently popular SIFT algorithm (Scale Invariant Feature Transform) conducted in-depth research and experimentation. The analysis of a large number of experimental results verify the algorithm to image the scale change, rotation changes, luminance variation remains unchanged on the viewing angle is changed, affine transformation, noise is also to maintain a certain robustness. SIFT algorithm has many advantages, but the need for multiple image scale space search the histogram weighted calculation of a large number of feature points, the image resolution in the face of time, registration time sometimes was geometric growth. Also retained a large number of SIFT feature points of the point of instability, will reduce the efficiency of the registration accuracy and registration. Based on the consideration of the above problems, the paper proposes a SIFT algorithm: taking into account the SIFT feature points is determined based on the feature point neighborhood pixel gradient, gradient large unique feature points better more image information precisely angle having a significant gradient change point is within the neighborhood window. So this will be Harris corner detection operator to introduce improved SIFT feature point extraction method based on SIFT algorithm image registration process, these feature points in the the multiscale spatial detection feature points while Harris corner extraction screened out more able to represent a feature point of the image information, but also reduces the amount of calculation performs vector description of the feature point. Experimental results show that a large number of not having a unique feature point is removed, retained the feature points are concentrated at the contour of the image reference more than doubled, the reduced number of feature points, but these points just is more representative of image information characterized in point. Greatly reduced due to the reduction of the feature points, the amount of calculation of the subsequent steps, to improve the speed of alignment. After the analysis of the experimental data proved that the improved algorithm for image registration, after the many advantages to retain the original SIFT algorithm also improves the speed of the alignment and correct matching probability. Finally, the prospect of the direction of the field of image registration require further study in the future.

Related Dissertations

  1. Research of Image Mosaic Technology,TP391.41
  2. Study on Image Registration and Image Matching of the Robot System for Substation Equipment Inspection,TP242.62
  3. Study on the Image Registeration Integrated with the LQ Model,R815
  4. Research and Implementation of Non-rigid Medical Image Registration Method Based on Improved Optical Flow Model,TP391.41
  5. Algorithm and Implementation Based on Non-rigid Medical Image Registration,TP391.41
  6. The Research of Phishing Detection Technology Based on Nearest Neighbor and Similarity Measurement,TP393.08
  7. GPU-based SIFT algorithm,TP391.41
  8. Real-time video display system key technology research,TP391.41
  9. Object Detection and Accurate Localization in Industrial Application,TP391.41
  10. Research for Super-resolution Reconstruction Technologies of Image Based on POCS Algorithm,TP391.41
  11. Medical image registration platform and combines new gray with geometric information Registration Measures,TP391.41
  12. Infrared and visible image registration and fusion research,TP391.41
  13. Coaxial different field picture image registration algorithm,TP391.41
  14. Feature points based spatial information distribution histogram matching method,TP391.41
  15. PDE-based image registration and fusion method,TP391.41
  16. Image registration based on the rotation angle calculation methods sand,TP391.41
  17. CT and ultrasound data registration studies and implementation,TP391.41
  18. The Next Generation Sequencing Data Processing,TP391.41
  19. Research on Ground-Based SFCW SAR Differential Interferometry Techniques,TN957.52
  20. Image Matching Based on Local Invariant Features,TP391.41
  21. Multi-spectral remote sensing image registration and fusion research,TP751

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net  Mobile